- Joined
- Jul 23, 2006
- Messages
- 176
- Reaction score
- 326
This is the first time it went below chance right? Still scary so far though, might be a stupid question but can one ask the gpt about its process for picking fights, like does it have some kind of technical analysis or narrative for how fights are likely to go based on past PBPs of the fighters?
It literally just uses news articles to form opinions if there are no statics to go off.I reckon it's lack of data on a bunch of rookies.
I imagine the story is the same next week.
These cards are too low level for accurate metrics.
It thinks every fight will end in a finish…interestingUFC Fight Night: Ribas vs. Namajunas
So OpenAI mygpts have been getting shitter and shitter. I have tried to recreate the gpt a bunch of time and after a few requests it starts to lose it instructions.
For this weeks pick I use Google Gemini 1.5 pro
You can try it here for free:
It’s probably my fault. Our robot overlord went 11-3 before I started putting money on its picks. I’ll take a crack at the picks That Gemini gave us just to see how they doUFC Fight Night: Ribas vs. Namajunas
So OpenAI mygpts have been getting shitter and shitter. I have tried to recreate the gpt a bunch of time and after a few requests it starts to lose it instructions.
For this weeks pick I use Google Gemini 1.5 pro
You can try it here for free:
Has there ever been a card with no decisions?UFC Fight Night: Ribas vs. Namajunas
So OpenAI mygpts have been getting shitter and shitter. I have tried to recreate the gpt a bunch of time and after a few requests it starts to lose it instructions.
For this weeks pick I use Google Gemini 1.5 pro
You can try it here for free:
There was a couple in the modern era I think but I can’t recall which ones.Has there ever been a card with no decisions?
Has there ever been a card with no decisions?
There was a couple in the modern era I think but I can’t recall which ones.
There was also UFC 76: Knockout which featured 0 knockouts and had three subs and the rest were decisions.
6 decisions on this card. After some research UFC Fight night 55 had all finishesI think the card where Yves Edwards KO'd Jeremy Stevens was an all finish card. I can't check right now though.
Below chance is not a good baseline for a sports-betting model though. It should be compared to the baseline of always picking the betting favorite IMO. That would probably get you close to 70% already. Even then, for actual betting it would be better if it spit out the chance of a fighter winning. Because then you could pick a strategy based on model's odds vs. bookmaker odds.This is the first time it went below chance right? Still scary so far though, might be a stupid question but can one ask the gpt about its process for picking fights, like does it have some kind of technical analysis or narrative for how fights are likely to go based on past PBPs of the fighters or does it just look at numbers?
Betting on the favourite works at in 64% for UFC cards in 2024 vs 68% for TSaL, but 78% before last two FNsBelow chance is not a good baseline for a sports-betting model though. It should be compared to the baseline of always picking the betting favorite IMO. That would probably get you close to 70% already. Even then, for actual betting it would be better if it spit out the chance of a fighter winning. Because then you could pick a strategy based on model's odds vs. bookmaker odds.
I still believe in you, this is a great idea,Welp that sucked. I think I will hold off on posting more predictions till I think I have a model that works better. Sorry guys I'll keep trying.